WHT.day MCP server

Each request carries a personal API key. Send it in the Authorization header, which is the preferred way:

Authorization: Bearer wht_YOUR_KEY

Some apps only let you paste a URL. For those, put the key in the path instead: https://wht.day/mcp/wht_YOUR_KEY. Both reach the same server.

• Creating a key. Sign in, open Settings, find API access, give the key a label such as “Claude”, and press Create API key. The key is shown once, together with an MCP link that has the key built in. Copy them then.
• Format. Keys start with wht_, followed by 43 random URL-safe characters.
• Who can create one. API access is part of a paid subscription, at $8 a month or $68 a year. The subscription is checked on every request: while it is inactive, keys stop working without being revoked, and they work again as soon as it resumes.
• What a key can reach. Only the archive of the account that created it, and only through the read-only tools below. A key can't sign in to the app or change any setting.
• Storage and revoking. WHT.day stores only a SHA-256 hash of each key, plus its first characters so you can tell keys apart. A lost key can't be shown again. Revoke it in API access and create a new one; a revoked key stops working on its next request.
• No OAuth. The server doesn't run an OAuth sign-in flow. The API key is the only credential.

Treat a key, and an MCP link with a key in it, like a password: anyone who has it can read your archive.

In each example, replace wht_YOUR_KEY with your key.

Claude Code

claude mcp add --transport http wht https://wht.day/mcp --header "Authorization: Bearer wht_YOUR_KEY"

Claude Desktop and claude.ai

Add a custom connector and paste the MCP link with your key built in: https://wht.day/mcp/wht_YOUR_KEY.

ChatGPT

With developer mode on, add the same MCP link as a connector. Or build a custom GPT and import https://wht.day/api/v1/openapi.json as an action. Choose API key authentication, type Bearer, and paste your key.

Cursor, VS Code and other clients

Any client that supports remote HTTP servers can connect with the endpoint and the header. The exact keys depend on the client; the usual shape is:

{
  "mcpServers": {
    "wht": {
      "type": "http",
      "url": "https://wht.day/mcp",
      "headers": {
        "Authorization": "Bearer wht_YOUR_KEY"
      }
    }
  }
}

Testing with curl

curl -s https://wht.day/mcp \
  -H "Authorization: Bearer wht_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

• Stateless. Each POST is handled on its own and answered with plain JSON. There are no MCP sessions and no event stream to resume.
• Dates. Dates are YYYY-MM-DD. A save belongs to the day you saved it, in your account's timezone, which get_profile reports.
• Excerpts and full text. Search and list results include a 600-character excerpt. Pass full: true or call get_save for the complete text. Every save carries its original url and an archiveUrl for the copy kept in WHT.day.
• Timelines use AI. Building a new timeline takes a few seconds, and the API builds at most 40 new timelines per account per day. Opening a timeline that already exists is unlimited.
• Instructions for the AI. When a client connects, the server tells it how to research the archive: find saves, read the ones it will rely on in full, use timelines for how a story developed, and cite the original url for every claim.

A 401 response looks like this:

{
  "jsonrpc": "2.0",
  "error": {
    "code": -32001,
    "message": "Missing or invalid WHT.day API key."
  },
  "id": null
}

A tool that can't do what was asked, for example because no save has that id or the daily timeline limit is reached, returns a normal result with isError: true and a plain-English message.